OpenAI Tests Persistent Mode for Codex That Keeps the AI Agent Working Indefinitely

OpenAI is developing a new “Persistent mode” for its AI coding agent Codex that would allow the agent to continue working autonomously until it is “put to sleep,” according to code reviewed by WIRED in August 2026.

Changes to Codex’s command line tool — which are public by default — show OpenAI has begun adding code for the new setting. The feature appears in Codex’s “reasoning effort” menu alongside existing modes that stop working on a task after a few minutes or hours, even if incomplete. Persistent mode appears to be one of the most computationally intensive options available. An OpenAI spokesperson confirmed the company is testing the feature but said there are no immediate plans to launch it broadly.

“OpenAI is a very bottom-up culture and many different things are explored on the open source repo which is a bit of our shared playground,” said Thibault Sottiaux, OpenAI’s head of core products.

Within Persistent mode, the code describes a feature called “proactivity.” Under this setting, the agent is instructed to create follow-up tasks for itself after completing a user’s request, work across sessions, draw on past user interactions, and message users unprompted — though it is told to do so sparingly. The code also sets limits: the agent is told Persistent mode does not expand its permissions, and that changes outside the user’s own system require user approval first.

The feature sits in the shared core of Codex rather than in terminal-specific code, suggesting it may eventually extend beyond the command line tool to products like the Codex desktop app and ChatGPT Work.

The development reflects OpenAI’s broader push to build AI agents with wider appeal. OpenAI, Anthropic, and Meta are all working on general-purpose agent products aimed at automating everyday tasks. Currently, AI agent users are largely software engineers. OpenAI CEO Sam Altman has publicly described his goal of turning ChatGPT into a proactive, always-on AI agent, which the company hopes will drive adoption of its most advanced models among a broader share of its user base.

Source: WIRED

This article was generated by AI and cites original sources.
Scroll to Top